Medical diagnosis, treatment, follow-up and research activities are nowadays strongly supported on different types of diagnostic images, whose main goal is to provide an useful exchange of medical knowledge. This multi-modal information needs to be processed in order to extract information exploitable within the context of a particular medical task. In despite of the relevance of these complementary sources of medical knowledge, medical images are rarely further processed in actual clinical practice, so the specialists take decisions only based in the raw data. A new trend in the development of medical image processing and analysis tools follows the idea of biologically-inspired methods, which resemble the performance of the human vision system. Visual attention models and sparse representations are examples of this tendency. Based on this, the aim of this doctoral proposal is to develop a set of computational methods for automatic morphometrical analysis, which combines the relevant region extraction power of visual attention models with the incorporation of a priori information capabilities of sparse representations.